Design And Implementation Of A Lidar-based SLAM Algorithm For Unmanned Vehicles

With the booming of "Industry 4.0" and "Internet +",various types of new technologies are being applied to the mobile robot industry.As an important branch of mobile robots,unmanned vehicles have attracted widespread attention.How to realize environment perception and self-positioning is one of the core issues in the field of unmanned vehicles.This paper comes from a practical project.When multi-unmanned vehicles perform tasks in industrial parks and indoor environments without GPS signals,they need to know their own pose information and environmental information.This paper will design positioning and mapping algorithm based on lidar for this purpose.In order to obtain unmanned vehicle's global position and real-time environmental information,and at the same time to guarantee the positioning and map building is real-time and accurate,this paper puts forward solutions based on laser radar,with a laser radar,and inertial measurement unit build global map of point cloud,and optimize the point cloud map.According to the real-time point cloud with point cloud map matching,we can solve unmanned vehicle real-time position,and obtain a real-time raster map which is constructed to facilitate path.The research of this paper mainly includes the following aspects:First,analyze the precision,real-time and robustness requirements of the project,design the hardware and software scheme of the unmanned vehicle,select and install components according to the requirements,and establish the coordinate system.Design the software operating system with ROS framework as the main body.Define the algorithm modules such as laser data preprocessing,point cloud map establishment,dynamic point cloud culling,real-time positioning and raster map construction.Second,an adaptive voxel filter was designed to simplify the point cloud.Through front-end matching and global optimization,the precise motion pose of the unmanned vehicle was obtained.Based on this,the point cloud data was inserted into the map to construct the global point cloud map.In order to deal with the residual shadows of dynamic objects in the point cloud map,a dynamic point cloud culling algorithm based on distance map was designed to eliminate the ghosting of dynamic point cloud in the map.Then,in view of the traditional ICP precision,poor real-time performance and robustness problems,adopt GICP point cloud matching to solve unmanned vehicle realtime position,improve the robustness and accuracy of the algorithm.To further improve the real-time and accuracy,design a Map segmentation algorithm which is based on KD tree.The map segmentation algorithm make match GICP can quickly find and real-time point cloud is most close to the local map,so we can obtain better initial value and improve the accuracy and real-time performance of the algorithm.The environment perception is realized by drawing global raster map and real-time local map.To improve the accuracy of global raster map,CSF algorithm is used to segment ground point cloud.In order to improve the real-time performance and accuracy of real-time raster map rendering,an improved RANSAC algorithm is designed to segment ground point cloud in real-time.Last,the algorithm was transplanted to the unmanned vehicle platform,and the experimental site was selected according to the engineering requirements to verify the positioning and mapping.Experimental results show that the accuracy,real-time performance and robustness of the algorithm meet the requirements.